In short
The comparison that matters is RF + VQC against PCA + VQC at the same number of qubits. Three classical models are shown only for context.
Four baselines were used: a traditional PCA + VQC, Random Forest only, a linear SVM and finally an RBF SVM. The linear SVM, RBF SVM and Random Forest are only for context and are not the bar to beat. This research’s purpose was to compare RF + VQC to PCA + VQC at equal qubits.
The traditional PCA + VQC took all 262,144 features and used centred PCA to get components; finally ZZFeatureMap encoding was used to encode a VQC. This was done at the same qubit counts as the RF + VQC methods.
The standalone RF used the same 100-tree forest and made classifications directly without a VQC, to show how much of the results come from the RF by itself.
The linear SVM gives a classical reference for how a current solution might perform, using LinearSVC with on all standardised features. For a classical non-linear reference an RBF SVM was used, with and gamma = “scale”, on all standardised features.